The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, specifically project files and metadata fields like context and operationGuidance returned by the openspec CLI. This creates a surface for indirect prompt injection where instructions embedded in project artifacts could attempt to override the agent's behavior.
Ingestion points: The skill reads files identified in the contextFiles array and parses JSON output from the openspec CLI within SKILL.md.
Boundary markers: The skill contains explicit instructions to treat external guidance as secondary to the built-in workflow. It specifies that these inputs are not evidence of task completion and cannot be used to bypass blocked states.
Capability inventory: The skill is authorized to perform file read/write operations and execute commands via the openspec CLI tool.
Sanitization: There are directives to identify and report conflicts between external context and the controlling workflow inputs, ensuring that the agent does not blindly follow external guidance that contradicts the system state.
[COMMAND_EXECUTION]: The skill utilizes the openspec CLI to manage project state, including listing changes, checking status, and fetching instructions. These operations are scoped to the openspec tool as defined in the skill's allowed tools.

openspec-apply-change